1
YouTube Science Channel Video Presenters and Comments: Female
Friendly or Vestiges of Sexism?
1
Mike Thelwall, Amalia Mas-Bleda. Statistical Cybermetrics Research Group, University of
Wolverhampton, UK.
Abstract
Purpose: This paper analyses popular YouTube science video channels for evidence of
attractiveness to a female audience.
Design/methodology/approach: The influence of presenter gender and commenter
sentiment towards males and females is investigated for 50 YouTube science channels with
a combined view-count approaching ten billion. This is cross-referenced with commenter
gender as a proxy for audience gender.
Findings: The ratio of male to female commenters varies between 1 and 39 to 1, but the low
proportions of females seem to be due to the topic or presentation style rather than the
gender of the presenter or the attitudes of the commenters. Although male commenters
were more hostile to other males than to females, a few posted inappropriate sexual
references that may alienate females.
Research limitations: Comments reflect a tiny and biased sample of YouTube science
channel viewers and so their analysis provides weak evidence.
Practical implications: Sexist behaviour in YouTube commenting needs to be combatted but
the data suggests that gender balance in online science presenters should not be the
primary concern of channel owners.
Originality/value: This is the largest scale analysis of gender in YouTube science
communication.
1. Introduction
Women are underrepresented in science. In almost all countries in the world, there are
more publishing male scientists, with proportions varying by field. This underrepresentation
is continuing despite progress in recent years and its causes are unclear (Sugimoto,
Larivière, Ni, Gingras, & Cronin, 2013). Contributory or associating factors include lower
female respect for science, fewer female scientist role models, poor pedagogy in science
classes, sexist course materials, cultural pressure (Blickenstaff, 2005) and gender
stereotypes (Ceci, Williams, & Barnett, 2009; Miller, Eagly, & Linn, 2015; Smyth & Nosek,
2015). More generally, females are underrepresented in STEM (Science, Technology,
Engineering and Maths) disciplines (Cesarsky & Walker, 2010; Ivie & Tesfaye, 2012; Kirkup,
Zalevski, Maruyama, & Batool, 2010; National Science Foundation, 2017). In quantitative
fields, continuing gender differences in the USA are not caused by biases against women
within academia; instead the socially constrained choices made by women seem to explain
differing career outcomes (Ceci & Williams, 2011). For example, young female biological
scientists may be less focused on authoring publications, damaging their long term
academic career prospects (Feldon, Peugh, Maher, Roksa, & Tofel-Grehl, 2017). It is
therefore important to understand the social context in which women choose science-
1
1. Thelwall, M. & Mas-Bleda, A. (in press). YouTube science channel video presenters and comments:
Female friendly or vestiges of sexism? Aslib Journal of Information Management.
2
related careers and their decisions at the start of these careers. This may reveal some ways
in which they are alienated from research.
The internet and YouTube are obvious choices for investigating gender issues in
science education. YouTube contains many different types of science-related videos,
including many that are documentary, recreational and educational (Erviti & Stengler, 2016;
Muñoz Morcillo, Czurda, & Trotha, 2016). It is widely used in school classrooms and by
university students to support learning (e.g., Barry, Marzouk, Chulak‐Oglu, Bennett, Tierney,
& O'Keeffe, 2016; Tan & Pearce, 2012) as well as for leisure-time explorations of science
related content, such as by watching TED Talks videos (see below) or science-related music
videos (Allgaier, 2013). It is also used as a research source (Kousha, Thelwall, & Abdoli,
2012).
Although the provision of free, high quality science content on the world’s second
most popular website YouTube (www.alexa.com/siteinfo/youtube.com on 9 June 2017) is a
societal benefit, it is concerning from a women’s empowerment perspective because
YouTube is a male-dominated corner of the internet. It has been the site of misogynist
abuse (Jane, 2014; Mourey, 2015; Wotanis & McMillan, 2014) and inappropriate personal
comments (Molyneaux, O’Donnell, Gibson, & Singer, 2008), even though positivity is more
common (Thelwall, Sud, & Vis, 2012). In male-dominated online spaces, gendered abuse and
stereotyping can thrive and become normalised so that females must try to cope with it or
combat it (Nardi, 2010). For example, a comparison of two high profile successful YouTube
comedians found that the woman was more criticised and subjected to more personal
comments (Wotanis & McMillan, 2014). Despite this, YouTube has seen the emergence of
more gender-inclusive cultures (Morris & Anderson, 2015) and so it is not clear that science
channels, if male dominated, would be unwelcoming for female viewers.
Gender is a factor in the popularity of YouTube science-related channels.
Professionally produced YouTube science videos seem to be more popular if they have a
male presenter, although the same is not true for amateur content and it is not known
whether the popularity is due to an increased male or female audience (Welbourne &
Grant, 2016). For TED Talks, male-presented videos are more popular (Sugimoto, Thelwall,
Larivière, Tsou, Mongeon, & Macaluso, 2013) but female presenters are more likely to elicit
positive or negative comments (Tsou, Thelwall, Mongeon, & Sugimoto, 2014). For the Khan
Academy YouTube science channel, 80% of commenters are male (Saurabh & Sairam, 2013).
Unless this is a special case or commenters are a highly gender-biased audience sample, it
seems that the YouTube audience for science videos is primarily male. In other genres, such
as TV, male presenters may also be more popular with female viewers (Sánchez Olmos &
Hidalgo Marí, 2016).
The predominance of males in some areas of science and YouTube raises the
possibility that hostile language (Alonzo & Aiken, 2004; Kayany, 1998; Lapidot-Lefler &
Barak, 2012; Moor, Heuvelman, & Verleur, 2010) may alienate female science channel
viewers. It tends to originate from males (Alonzo & Aiken, 2004) and is not necessarily
related to the content of a video (Lange, 2007). Males on YouTube are more likely to
comment on the attractiveness of vloggers (Molyneaux, O’Donnell, Gibson, & Singer, 2008),
and prominent female YouTubers are routinely forced to deal with threatening sexist abuse
(Mourey, 2015). Offline, male sexual humour is used to relieve anxieties about masculinity
(O’Connor, Ford, & Banos, in press; Pascoe, 2013). In this context, commenters may
perceive inappropriate sexual references as being humorous and inoffensive. This would be
a mistake because, for example, the occasional “low level” sexist behaviour (or
3
microagression) that is a fact of life for some women in physics and astronomy has tangible
impacts. These include the consequent social pressure on females to manage their
appearance to be perceived as serious and intelligent by their colleagues (Barthelemy,
McCormick, & Henderson, 2016).
From the above review, males are likely to dominate the presenters and viewers of
YouTube science videos, potentially creating an unwelcome space for female viewers.
Nonetheless, no previous study has sought evidence of the reasons for gender imbalances
on YouTube science videos or attempted to provide recommendations for attracting a wider
audience. The current paper addresses this gap by comparing the gender ratios of the
audiences of a set of popular science channels (RQ1). It also seeks evidence of an alienating
environment for women by male presenters or in the sentiments expressed towards
females in the comments left underneath the videos (primarily RQ2b). This is driven by the
following research questions.
RQ1: Are females less likely to watch YouTube science channels that have male
presenters?
RQ2a (MF+<FF+): Are male science video commenters less positive than female
science video commenters when discussing females?
RQ2b (MF->FF-): Are male science video commenters more negative than female
science video commenters when discussing females?
RQ2c (MM+<FM+): Are male science video commenters less positive than female
science video commenters when discussing males?
RQ2d (MM->FM-): Are male science video commenters more negative than female
science video commenters when discussing males?
2. Methods
The overall research design was to obtain a large sample of popular YouTube science
channels to investigate the influence of presenter gender on the ratio of male to female
commenters (RQ1) and to look for evidence of hostility towards women in their comments
(RQ2). This is a novel approach that could be contrasted with more exploratory strategies
for YouTube comment analysis (e.g., Thelwall, in press-a).
2.1 YouTube science channels
There are many different science channels on YouTube and so a method was needed to
obtain a definitive list. A YouTube channel search for the keyword Science yielded
11,192,130 channels, including some, like Holy Fucking Science, that emphasise
entertainment. Web searches were therefore used instead to identify recommended lists of
varied but high quality science channels. The best list found was that of the GeekWrapped
science gadget website https://www.geekwrapped.com/posts/youtube-science-rockstars-
shows. Whilst this list is from a commercial site rather than a reputable source, all channels
are popular and contain high quality science content. The use of a specific list is important
for increased objectivity in comparison to a manually generated list. A manually-created list
would be the result of subjective decisions made by the research team that might
subconsciously be affected by the research goals. Such a list could also be accused of being
selected to demonstrate the research goals. The first fifty channels from the pre-existing list
were used as the raw data for this paper, except that two were lists rather than channels
and were replaced by the 51
st
and 52
nd
channels.
4
2.2 Channel information, presenters, commenters and comments
The list of videos in each channel and the comments on these videos were downloaded
using the YouTube API 5-8 June 2017 in the free software Mozdeh
(http://mozdeh.wlv.ac.uk). For each channel, only one comment was allowed per user (the
most recent one on the most recent video) to prevent individual prolific commenters from
influencing the results. For videos with many comments, YouTube returns the most recent
about 350.
The gender of each commenter was inferred from their username. When possible
(either through spaces or camel case) usernames were split into multiple parts. If the first
part matched a name that was used at least 90% by males or females in the US census (e.g.,
see: Sugimoto, Larivière, Ni, Gingras, & Cronin, 2013) then the commenter was assigned
that gender. First parts of Mr, Mrs, Ms and Miss were also assigned to the appropriate
gender. Most usernames did not match these rules and were left unassigned. For example,
only 35% of Tyler DeWitt and 23% of Explorium commenters were assigned a gender. From
manual checks of the results in the current and previous projects, this process seems to
have an accuracy level of considerably above 90% in terms of the gender projected by the
name, if not the (unknown) gender of the user. The only potentially incorrect classification
found in the manual checks was Hui Yang (assigned as female). Whilst Hui is more common
for females, at least in the U.S. 1990 census, it can also be used by males. The name-based
gender identification procedure will generate some false matches and does not work for
transgender individuals but can identify a predominantly male group and a predominantly
female group. A US source was chosen for the name list because the USA is the largest user
of YouTube, is a multi-cultural nation, and has an informal naming tradition that captures
many shortened name forms (e.g., Lizzie). It is not possible to check whether the method
has a greater success rate for one gender, biasing the results, because most of the
unassigned usernames are gender neutral (e.g., names like Newb33, CouscousLover).
Nevertheless, any bias seems likely to be constant between channels so the main fact that it
may influence is the overall proportion of female commenters.
Commenting on a YouTube video is a way to interact with its creator or other users.
Many comments are factual or short statements but some address other people by name or
with a pronoun. Gendered pronouns were used as a universal method to identify that a
comment was referring to a male or female. Comments matching the query he his him man
boy himself -she -her -woman -girl -herself were assumed to be comments to or about a
male and comments matching the query she her woman girl herself -he -his -him -man -boy -
himself were assumed to be about a female. These are heuristics because people may be
referred to by name (e.g., Mary, Nick) but the advantage of pronouns is that they suggest a
deeper involvement in the person referred to by the fact that they do not need to be
individually named, or are discussed multiple times so that they do not need to be named
every time that they are referred to in a comment.
Commenter gender information was combined with pronoun queries to generate
four separate sets of comments for each channel, each containing at most one comment
from each user.
MM: Male-authored comments containing exclusively male pronouns.
MF: Male-authored comments containing exclusively female pronouns.
FM: Female-authored comments containing exclusively male pronouns.
FF: Female -authored comments containing exclusively female pronouns.
5
For some channels, there were few or no comments in the FF category and so the data set
for the second research question was restricted to the 32 videos with the most comments.
This gave a simple cut-off since the 33
rd
channel had no FF comments.
2.3 Sentiment towards presenters in comments
The strength of positive and negative sentiment in each comment in the MM, MF, FM, and
FF groups was identified with the software SentiStrength (sentistrength.wlv.ac.uk) that
exploits a lexicon of sentiment terms in addition to a set of linguistic rules (e.g., for
negation, idioms and booster words) to estimate the strength of positivity and negativity in
a text. It assigns a score of 1 (no positivity) to 5 (very strong positivity) and a second,
independent score of 1 (no negativity) to 5 (very strong negativity) to each text. For
example, the comment, “Great point about pi!” would score 4 for positivity because of the
word great, which is in SentiStrength’s lexicon with a default score of +3, and the
exclamation mark, which boosts the strength of the positive sentiment by 1. It scores -1 for
negativity, indicating no negative sentiment (zeros are not used). Lexical software that uses
a pre-defined list of sentiment terms and additional linguistic rules (Taboada, Brooke,
Tofiloski, Voll, & Stede, 2011) like SentiStrength is preferable to machine learning (Pang &
Lee, 2008) for social science research purposes because the latter can detect controversial
topics as proxies for sentiment (Thelwall, Buckley, & Paltoglou, 2012). SentiStrength was
chosen for accuracy approaching human-level on YouTube comments (as found by
comparisons between its results and three human coders for a random set of YouTube
comments: Thelwall, Buckley, & Paltoglou, 2012) as well as for its dual system that allows
negative sentiment to be analysed independently from positive sentiment, which is
important for the research goals. Sentiment analysis contains a small gender bias because
females tend to express sentiment more explicitly than males online (e.g., Thelwall, in press-
b) but this does not affect the current paper much because the main comparisons are
between commenters of the same gender, but different targets (MM vs. MF and FF vs. FM).
For each channel and each group (MM, MF, FM, FF), the average positive and
negative sentiment strengths of the comments were calculated separately. A 95%
confidence interval was calculated for each one using the standard normal distribution
formula. This is an approximation since the data is skewed (mode 1 in all cases) and discrete
rather than continuous. The data also violates the statistical independence assumption
because comments relating to the same video might be influenced by each other. The
confidence limits should therefore be interpreted as indicative estimates rather than robust
values. Because of this, and for simplicity of analysis of multiple results, differences in
average sentiment will be interpreted as significant when confidence intervals do not
overlap. This is a compensatory conservative approach because a small overlap between
confidence intervals is consistent with statistically significant differences (Schenker &
Gentleman, 2001).
3. Results
3.1 RQ1: Presenter gender
The popular science channels mostly had male or mixed presenters, with only a few female
presenters. In the mixed cases, males seemed to dominate numerically in all channels. The
presenter has varied degrees of prominence in the channels, from being the central visible
